Contrasting trajectories of morphological diversification on continents and islands in the Afrotropical white‐eye radiation

Morphological and lineage evolution are predicted to follow different patterns in island and mainland radiations. However, the extent to which these geographical contexts influence evolutionary trajectories remains poorly understood, in part because few studies have focused on species‐rich clades colonizing continents and archipelagos over comparable timeframes. Focusing on a diverse songbird clade radiating on the African continent and adjacent islands, we tested whether morphological evolution is best explained by adaptive or non‐adaptive processes, and whether mainland and island lineages evolved to occupy different regions or volumes of morphological space (morphospace).


| INTRODUC TI ON
Understanding evolutionary processes driving lineage and phenotypic diversification is a central concept in evolutionary biology, in which biologists have attempted to discern if there are commonalities across clades and regions (Harmon, Schulte, Larson, & Losos, 2003). In particular, recent evolutionary radiations containing a proliferation of taxa at lower taxonomic levels provide critical insight while avoiding complicated causal events of older radiations. A general feature attributed to insular (oceanic island, or lake) endemic radiations is that they are often characterized by adaptive radiation (Ford et al., 2016;Jønsson et al., 2012;Losos & Ricklefs, 2009;Lovette, Bermingham, & Ricklefs, 2001;Mahler, Revell, Glor, & Losos, 2011;Seehausen, 2006). Under an adaptive model, ecological opportunity (Schluter, 2000) is thought to drive the rapid increase in diversification during the early stages of the radiation, when there is less competition through fewer species and more available niches.
As the radiation progresses, competition between species intensifies and there are fewer available resources, resulting in decreasing opportunities for speciation and/or greater extinction of species (Gavrilets & Vose, 2005). The scenario of an 'early-burst and slowdown' (density dependence) model for adaptive radiation has led to a prediction that lineage and morphological diversification are correlated (e.g. Rabosky et al., 2013). However, despite an early burst of lineage diversification being identified as a general pattern of adaptive radiations, this signal does not necessarily correlate with an early burst of morphological evolution. Instead, there is evidence to suggest that a constrained model of evolution (Ornstein-Uhlenbeck [OU] -a random walk, but with a selective optima) is more commonly supported, even for classic adaptive radiation examples such as Darwin's Finches and Anolis lizards (Harmon et al., 2010), although it has been shown that these models may incorrectly be favoured over simpler (Brownian motion) models (Cooper, Thomas, Venditti, Meade, & Freckleton, 2016). Irrespective of this, reasons for the absence of an early-burst model of trait evolution are unclear.
Studies of adaptive radiations have primarily focused on insular systems, whereas most biodiversity occurs on continents where land area is far larger and often far more complex in terms of geography or topology (May, 1994). The increased age and species richness of continental biota increases the potential for interspecific competition, leading to the prediction that mainland taxa should evolve more slowly and in a more constrained manner (Schluter, 1988;Tobias et al., 2014;Webb, Ackerly, McPeek, & Donoghue, 2002).
Compared with island systems, continental radiations may be characterized by more allopatric diversification and reduced constraints on ecological niche divergence (Losos & Ricklefs, 2009), increasing the likelihood of non-adaptive or constrained models of diversification (Rundell & Price, 2009;Tobias et al., 2014). However, previous studies provide conflicting evidence regarding the dynamics of diversification in mainland clades. On the one hand, adaptive radiation is clearly not limited to insular contexts because several continental radiations fit an adaptive model of lineage and morphological diversification, for example, mainland Anolis lizards (Pinto, Mahler, Harmon, & Losos, 2008) and cichlid fishes (López-Fernández, Arbour, Winemiller, & Honeycutt, 2013). On the other hand, while many continental clades are characterized by a density-dependent model of lineage diversification (Phillimore & Price, 2008;Rabosky & Lovette, 2008), it is possible that this pattern is generated by standard models of allopatric speciation (Pigot, Phillimore, Owens, & Orme, 2010). Moreover, several studies focusing on continental tropical clades have revealed a near-constant rate of diversification, suggesting that these radiations may either be too young to have reached their ecological limit or are less restricted by ecological opportunity (Day et al., 2013;Derryberry et al., 2011;Liedtke et al., 2016). Several studies of continental avian radiations have concluded that evolutionary processes underlying morphological disparity may be uncoupled from those driving species diversity (Derryberry et al., 2011;Schweizer, Hertwig, & Seehausen, 2014).
For example, a constant rate of lineage diversification may be coupled with an adaptive signal of morphological diversification, although these contrasting signals may reflect the uncertainty of these models to infer correct processes.
Clades of highly dispersive organisms, such as those that can colonize multiple continents and can expand their ranges within extensive archipelagos, provide an opportunity to directly investigate evolutionary dynamics across landscapes of different spatial scales. However, there are limited studies investigating these trends within mainland and island settings (but see Schluter (1988), Pinto et al. (2008), Fritz, Jønsson, Fjeldså, andRahbek (2012)), with few specifically investigating both lineage and phenotypic evolution. Focusing on a classic example of adaptive radiation, Pinto et al. (2008) found lineage diversification across Caribbean island and Central and South America mainland Anolis lizards were comparable, identifying an adaptive signal, yet island species occupied different parts of morphospace, suggestive that competition and/or predation between these regions differed. However, it is not clear if this is typical in radiations spanning mainland and islands, particularly in groups that exhibit limited morphological change over time.
To address whether mainland and island lineages followed the same pattern of morphological and lineage evolution, we focus on the Afrotropics, which we define as continental Africa-Arabia, and associated island archipelagos including the western Indian Ocean, Madagascar and Gulf of Guinea. The advantage of this study region is the availability of a comprehensive species-level taxonomy of Zosterops (Martins et al., 2020) spanning a similar number of taxa distributed on the continent versus surrounding islands. Based on new measurements of nine morphological traits, we used evolutionary models to test if morphological diversity of Afrotropical Zosterops (n = 46 species and candidate species), and its constituent mainland and island radiations, is best explained by adaptive or non-adaptive processes. We also tested whether lineage diversification in these different geographical contexts followed a pattern of density dependence and mapped traits of all study species in multivariate morphological space (morphospace) to assess whether mainland and island lineages evolved to occupy different regions of morphospace. Finally, to place our findings in wider context, we compiled morphological data for the entire zosteropid clade, including both Zosterops and related genera, and compared how Afrotropical radiations have explored morphospace in relation to taxa from the rest of the Palaeotropics (n = 75 species).

| Study system
The white-eye family (Zosteropidae) is one of seven families in the diverse Old World babbler radiation (Cai et al., 2019). Members of Zosteropidae are small (ca. 10-12 cm) arboreal passerine birds, often gregarious, and regarded as ecological generalists, with reported diets consisting of a mixture of insects, fruits and nectar (van Balen, 2019). The geographical distribution of the clade extends across the Palaeotropics, which includes the Afrotropics and Austral-Asia (southern and eastern Asia, Australasia, and the eastern Indian Ocean islands) (van Balen, 2019). A few Zosterops species have colonized the temperate zone as far as South Africa, Japan and New Zealand. Within the Afrotropics, the division between island and continental forms is roughly equal (Martins et al., 2020), whereas outside the Afrotropics the majority of species are insular forms with another centre of diversification in continental South and East Asia. Despite exhibiting a signal of density dependence (Moyle et al., 2009), Zosterops are remarkably homogeneous in morphology and plumage (Mayr 1963;Mees, 1953;Moreau, 1957), apart from some aberrant island forms (Melo et al., 2011).

| Molecular dated tree
We used the dated comprehensive phylogenetic tree from Martins et al. (2020) based on three mitochondrial markers (Cytb, ATP6 and ND3), five nuclear introns (TGFß2, G3PDH, Fib 7, CHD1 and MUSK) and comprehensive sampling of all known Afrotropical species and subspecies (with the exception of Z. maderaspatanus menaiensis). Overall, this timetree is based on results from phylogenetic inference and quantitative species delimitation methods for 46 Afrotropical taxa (described species, and possible candidate species [represented by trinomial nomenclature]), comprising 25 mainland and 21 island taxa, respectively (for locality data, see Martins et al., 2020, Appendix S1).

| Collection of trait data
We collected trait data from museum specimens of nearly all sampled Zosterops taxa (n = 45 species) included in the dated phylogenetic analysis of Martins et al. (2020), with only the extinct species Z. semiflavus missing. In addition, 59 Zosterops species and 16 species in other Zosteropidae genera were included from the remainder of the Palaeotropics, totalling 120 species (508 individuals) (see Appendix S1, Table S1 for biometric data). Where available, we selected approximately two males and two females per species collected from the same taxon. We selected ecomorphological traits linked to the ecological niche, with methods following Trisos, Petchey, and Tobias  (Pigot et al., 2020). To test for repeatability of measurements between recorders, FCM independently measured all traits from 12 specimens (across five species) originally measured in the JAT dataset to ensure congruence between observers. Additionally, a random sample of 11 specimens (different taxa) was remeasured by FCM in order to demonstrate intra-observer reliability. An ANOVA was applied to both datasets and showed that while measurements varied between traits (as expected), there was no evidence for any differences in the observers (i.e. FCM versus. JAT), and no interaction between observer and trait. Previous analyses on much larger samples of species found similar results, suggesting that measurement bias among observers is too low to influence results of macroecological studies across multiple species (McEntee, Tobias, Sheard, & Burleigh, 2018;Pigot et al., 2020).

| Morphometric analyses
Principal component analyses (PCA) were initially conducted using only those taxa included in the phylogeny (excluding Z. semiflavus and Austral-Asian outgroups), in which Africa-Arabia mainland taxa were based on species delimitation analysis and recommendations from Martins et al. (2020). Data were log-transformed, using the natural logarithm, to reduce variance between taxa (Ricklefs & Travis, 1980). For each trait, we calculated the mean for each taxon. Although sample size is fairly limited (ca. 4 individuals measured per taxon), potentially resulting in biases, small samples have been shown to provide relatively accurate estimates of mean size when comparing across species (see Cardini & Elton, 2007). To determine if these data needed to be phylogenetically corrected, the log-transformed data were analysed using a phylogenetic size correction via generalized least squares regression (Revell, 2009) in 'phytools' v0.6-99 (Revell, 2012), R 3.4.3 (R Development Core Team, 2018). As this was non-significant (p > 0.5), the PCA loadings of the non-phylogenetically corrected data were extracted using past 3.18 (Hammer, Harper, & Ryan, 2001) to obtain information on the proportion of the contribution of each parameter for each PC (see Appendix S1, Table S2 for PCA scores). Together, the first four PCs of the Afrotropical radiation explained 89.98% of the variance (52.94%, 20.72%, 10.84% and 5.49%, respectively) and these axes were used for subsequent comparative evolutionary analysis. To compare morphospace occupation of taxa comprising the Afrotropical radiation to Austral-Asian taxa, we included a further 59 Zosterops species from southern and eastern Asia, and 16 Zosteropidae from other genera using the methods previously described. Here, PC1-4 explained 93.07% of the variance (65.96%, 16.43%, 6.27%, 4.41%, respectively) based on the past analysis. A 95% confidence interval ellipsis was included for comparisons of identified groups for these axes.

| Comparative evolutionary methods
To test if morphological evolution shows a departure from the null model of Brownian motion, we used Disparity-through-time (DTT) analyses (Harmon et al., 2003). We employed the recently developed rank envelope method to compare the empirical data to the null model simulations, following Murrell (2018). Here, the DTT curves from the null model simulations are ordered according to their most extreme relative disparity value, and the empirical DTT curve is then ranked to see if it falls within the most extreme 5% of curves in which case the null hypothesis of Brownian evolution is rejected. This method has been shown to generally have higher power to detect non-Brownian evolution than other popular DTT methods, and it is straightforward to visually inspect the time points where the null model is rejected (Murrell, 2018). Since the method gives a rank to each simulation and empirical DTT curve based upon their most extreme relative disparity value, ties in ranking between empirical and simulation curves are almost certain to occur. Rather than break ties, we report a p value interval, within which the true p value is estimated to be found, and large numbers of simulations ensure this interval is small. We ran all models with 5,000 simulations. The scores from the PCA were extracted for each taxon for principal component axes 1-4 and combined with the timetree. DTT plots that show high subclade disparity, in which there is greater disparity within subclades, have been described as evidence of convergent evolution. However, caution in interpretation is required since, assuming the phylogenetic tree is ultrametric, high subclade disparity (i.e. disparity >1) can also be caused by an accelerating rate of trait evolution, and it is not possible to distinguish between this model and the OU model with a single optimum using DTT analysis (Uyeda, Caetano, & Pennell, 2015). Conversely, plots that have low subclade disparity (i.e. disparity <1), where there is greater disparity between subclades, might indicate a signal of adaptive radiation (Harmon et al., 2003). Building on Murrell (2018), we analysed all PC axes separately but within one test (i.e. one p value interval is returned), with visual inspection of the curves against the null model confidence interval showing which PCs are responsible for rejecting the null model.
Models were compared using the Akaike Information Criterion (AIC) in which a lower AIC score indicates the model that is best fitted by the data. To account for sample size, we also calculated the AICc, which was used to obtain the deltaAICc and Akaike weights (Aw). All analyses were conducted in R 3.4.3 (R Development Core Team, 2018).

| Diversification rate analyses
We used several methods to investigate lineage diversification. To detect if there was an early burst followed by a slowdown in diversification rate over time, we calculated the ϒ statistic of Pybus and Harvey (2000), and generated a lineage-through-time (LTT) plot using 'ape' software (Paradis et al. 2004). We also applied a likelihood approach (Morlon, 2014) and compared two constant rate models -(1) pure birth (PD) and (2) birth-death (BD) -to four variable rate models: (1) PB with an exponential speciation rate; (2) BD with a constant speciation rate and exponential extinction rate; (3) BD with an exponential speciation rate and constant extinction rate and (4) BD with both exponential speciation and extinction rates, where rates are changing over time, in rpanda package v1.3 (Morlon, 2014). Model fit was compared using AIC and adjusted for small sample sizes (AICc), which was used to obtain Aw.
These analyses were conducted in R 3.4.3 (R Development Core Team, 2018). To test if subclades of the Afrotropical radiation diversified under distinct rate regimes, we also implemented bamm 2.5.0 (Bayesian Analysis of Macroevolutionary Mixtures, Rabosky (2014)) and used bamm tools version 2.1.6 (Rabosky et al., 2014) to investigate rate dynamics. The following priors placed on the evolutionary rate parameters, which are critical to the performance of bamm, were generated for our dataset using bamm tools (expected number of shifts = 1; λ Init.Prior = 0.126; λ ShiftPrior = 0.578; μ Init.Prior = 0.126; λ TimeVariablePrior = 1) and applied to downstream analysis. We ran three independent MCMC runs for five million generations each sampling every 1,000th, and used tracer 1.7 (Rambaut, Drummond, Xie, Baele, & Suchard, 2018) to assess convergence. Models were compared in bamm tools using Bayes factors (BF).

| Dynamics of lineage diversification
Visual inspection of the LTT plot reveals a pattern of rapid cladogenesis followed by declining diversification rates for the Afrotropical radiation (Appendix S1, Figure S1.1), which is supported by a gamma (ϒ) statistic that is a strongly negative (−4.43, p ≤ .001 [two-tailed p value]). The variable rate model, pure birth model, with an exponential speciation rate was by far the best fit to our data (Table 1).
Strongly declining net diversification rates were also recovered from the bamm analysis (Figure 1) in which a mean speciation rate (λ) was 1.82 (1.32-2.48 upper/lower 95% highest probability density [HPD]), and the mean extinction rate (μ) was 0.54 (0.03-1.56 upper/ lower 95% highest probability density [HPD]). We found no strong evidence for any rate shift (BF = 0.86) among subclades (i.e. no rate heterogeneity), indicating that neither island nor mainland clades exhibited faster rates with respect to one another.

| Dynamics of trait evolution
The DTT analyses showed that when Afrotropical taxa are analysed together across PC1-4 disparity is predominately partitioned within subclades throughout the entire radiation (Figure 2a). Closer inspection reveals, PC1, PC3 and PC4 to all be significantly differ- TA B L E 1 Testing models of diversification for the Afrotropical Zosterops radiation (comprising 46 taxa). Models include pure-birth (PB); birth-death (BD); PB with an exponential rate (PBλexp); BD with a constant speciation rate and exponential extinction rate (BDλcst-μexp); BD with an exponential speciation rate and constant extinction rate (BDλexp-μcst); BD in which speciation and extinction rates are both exponential (BDλexp-μexp). Log-likelihood, AICc, ΔAIC (the difference in AICc with the best model) and Akaike Weights (AW) are shown. Bold text denotes the best model

| Distribution of mainland and island taxa in morphospace
Mainland Afrotropical taxa are tightly clustered in morphospace

| Contrasting patterns of morphological evolution between mainland and island white-eyes
Our analyses show that morphological evolution of Afrotropical zosteropids is best explained by convergence, and that constrained as opposed to adaptive evolutionary processes have occurred during their radiation. This finding contrasts with adaptive bursts of trait diversification found in classic avian adaptive radiations (see Grant & Grant, 2006;Price, 2008), as well as early adaptive signals shown in other spatially extensive radiations in birds occurring on continents (Derryberry et al., 2011;Schweizer et al., 2014). However, it supports previous studies showing a pattern of rapid evolution followed by stasis may be rarer than one of constraint on phenotypic evolution (Harmon et al., 2010), although see Cooper et al. (2016).

The absence of an adaptive signal of phenotypic evolution in
Zosterops is not unsurprising given that taxonomists (Mees, 1953;Moreau, 1957) have struggled to differentiate taxa, particularly those from the Africa-Arabian mainland, due to their overall remarkably cryptic phenotypes. In island Zosterops, the lack of an early adaptive signal perhaps reflects constraints on trophic specialization associated with their generalized diets, or else may suggest that this clade lacks the inherent developmental plasticity (i.e. 'evolvability') of beaks and foraging behaviours associated with other avian clades that underwent adaptive radiation on islands. In line with trophic F I G U R E 2 Disparity-Through-Time (DDT) plots for the Afrotropical zosteropid radiation generated using the rank envelope method. (a) including all mainland and island Afrotropical taxa; (b) island Afrotropical taxa only; (c) mainland Afrotropical taxa only, depicting the first four principal component axes. Disparity profiles are indicated by lines (PC1-4), average Brownian motion simulation by the dashed lines, and 95% confidence intervals from 5,000 simulations by the grey polygons. The x-axis is relative time in which zero is the start of the radiation [Colour figure can be viewed at wileyonlinelibrary.com]

(c)
Relative Time Relative Disparity traits, the diversification of plumage colouration in zosteropids is also subtle, although further study is warranted to investigate evolutionary patterns in some of the clade's more variable plumage features, including eye-ring size and belly colouration (Milá, Warren, Heeb, & Thébaud, 2010;Oatley, Voelker, Crowe, & Bowie, 2012).
Overall, since our study focuses on the exceptionally short timescale of the Afrotropical radiation, future analyses should consider the entire zosteropid clade, particularly as Austral-Asian lineages have diversified across a considerably larger volume of morphospace.
On closer inspection of the Afrotropical radiation, which included near even taxonomic richness of mainland versus island members, the signal of evolutionary convergence appears to have been largely driven by mainland taxa. Morphological diversification in this group of taxa strongly favoured an Ornstein-Uhlenbeck model, indicating bounded evolution around a phenotypic optimum. Constrained phenotypic evolution in cryptic continental species has also been shown in agamid lizards (Smith, Harmon, Shoo, & Melville, 2011). The significant recent increase in trait disparity of mainland Zosterops taxa is potentially explained by repeated patterns of niche divergence driven by Plio-Pleistocene climatic fluctuations (Cox, Prŷs-Jones, Habel, Amakobe, & Day, 2014). These authors investigated East African montane Zosterops taxa using phylogenetic inference and showed a general trend for sister species to comprise highland and lowland members. The highland birds, which tend to have more restricted ranges on mountain tops, are larger and heavier, whereas lowland birds are smaller with larger ranges (Habel et al., 2015).
Although mainland species are similar and only show small phenotypic differences, adhering to the typical Zosterops phenotype, there is some evidence from our study that highland and lowland bauplans, which show some separation (Figure 4a), may have evolved rapidly based on the timetree (Figure 1). Such rapid phenotypic disparity has also been found in highland and lowland island Z. borbonicus populations from Réunion (Cornuault et al., 2015), as well as in island populations of Z. lateralis which show phenotypic differentiation from mainland source populations (Australia) even in populations founded in the southwestern Pacific over the last 200 years .
When comparing PC1 of traits ( between these highland and lowland mainland species (Figure 4a).
However, when mainland birds are compared to island taxa, we find that the volume of morphospace occupied by mainland birds is relatively restricted (Figure 3). It appears that repeated diversification into highland and lowland continental habitats has been followed by phenotypic convergence towards particular optimal trait combinations. In addition, some speciation events appear to be driven by vicariance rather than colonization of different habitats since some sister species of continental Zosterops are comprised of both lowland or both highland forms (Martins et al., 2020). Moreover, we only found small variations within species from each of these habitats, suggesting that morphological novelty is constrained in continental birds. This may ultimately be due to fewer ecological opportunities since continental communities are probably closer to carrying capacity and the available niches more highly saturated with potential competitors (Losos & Ricklefs, 2009;Tobias et al., 2014).
In contrast to the mainland taxa, island Zosterops show a relatively weak signal of morphological convergence, in which phenotypic diversification is typically not distinguishable from a random-walk pattern (Brownian motion, BM). Although a BM model of character evolution may occur when evolutionary change is neutral, in which traits change only as a result of genetic drift (Lande, 1976), it can also occur from natural selection (Harmon, 2018). Certainly, island evolution has been shown to result from neutral processes, such as genetic drift and founder events (González-Quevedo, Spurgin, Illera, & Richardson, 2015;Spurgin, Illera, Jorgensen, Dawson, & Richardson, 2014), and divergent natural selection (Clegg, Frentiu, Kikkawa, Tavecchia, & Owens, 2008;Grant & Grant, 2006;Losos & Ricklefs, 2009;Milá et al., 2010;Ryan, Bloomer, Moloney, Grant, & Delport, 2007). As these mechanisms are not mutually exclusive, it is likely that both neutral and selective factors underlie the evolutionary history of island zosteropids, as has been shown in island populations of Z. borbonicus (Milá et al., 2010).
One reason why island Zosterops lineages may show less evidence of evolutionary constraints is that they experience reduced levels of interspecific competition. Most islands are occupied by single endemic taxa, and those occupied by two (e.g. Réunion, Mauritius), are the result of independent invasions (van Balen, 2019). Single colonization events have often led to Zosterops radiations across archipelagos, for example, the 'Indian Ocean maderaspatanus' and 'Ancient Indian Ocean' clades diversified in the western Indian Ocean islands, but in general they involve a single species per island (Warren et al., 2006).

Interspecific competition may therefore be reduced in island
Zosterops, not only because they co-occur with no or few close relatives but also because the total number of sympatric passerine bird species on islands is also much lower than on the mainland. Thus, an ecological release may have allowed for greater evolutionary novelty in these island radiations, as implied by their exploration of a greater volume of morphospace compared to mainland birds (Figure 3).  (Lovette et al., 2001), Hawaiian honeycreepers (Lerner, Meyer, James, Hofreiter, & Fleischer, 2011) and Madagascan vangas (Reddy, Driskell, Rabosky, Hackett, & Schulenberg, 2012). However, island forms often occupy different areas of morphospace from their mainland relatives, as shown in Hawaiian honeycreepers (Gavrilets & Losos, 2009) and Caribbean anoles (Pinto et al., 2008)

| Evidence for density-dependent lineage diversification
The pattern of phenotypic evolution contrasts with the early burst and subsequent slow-down pattern of lineage diversification identified by our analyses, in line with a previous study (Moyle et al., 2009). These authors investigated a subset of Zosterops species from across the radiation (ca. 30 species) and also showed exceptional rates of lineage accumulation relative to Zosteropidae as a whole, suggesting dispersal ability, generalist ecology, short generation time, sociality and capacity for phenotypic evolution as possible influencing factors. Although the early burst pattern of lineage accumulation has been suggested to indicate adaptive radiation (Gavrilets & Vose, 2005;Rabosky & Lovette, 2008), this pattern can also be recovered from simple geographical speciation in the absence of niche filling mechanisms (Pigot et al., 2010). Given the apparent lack of ecological diversification, and limited insular diversification It has been suggested that island clades diversify more rapidly than mainland clades owing to reduced gene flow and the availability of new niches (e.g. Grant & Grant, 2006;Price, 2008). However, we found no evidence for this in Afrotropical zosteropids since rate shifts were not identified from our analyses (Figure 1a), perhaps due to relatively small sample size. In general, the comparable rates across island and mainland clades that we identify in zosteropids supports the findings of Pinto et al. (2008) who suggested that recent, rapid diversification of anoles proceeded in similar ways in insular and continental landscapes. This is contrary to a global avian study that showed bird species with more than half of their range on islands have higher diversification rates than mainland bird species (Jetz et al., 2012). Clearly, further investigation is needed as to why these different patterns have emerged in clades of highly dispersive organisms that have colonized multiple continents and archipelagos.

| CON CLUS ION
Our results suggest that continental lineages in the Afrotropical Zosterops radiation follow different evolutionary trajectories to island lineages. Unlike some widespread continental avian clades that show an early burst of phenotypic evolution (Derryberry et al., 2011;Schweizer et al., 2014), mainland zosteropids show constrained evolution, with apparent convergence on particular morphological phenotypes. The lack of density dependence of morphological traits in mainland forms is perhaps not surprising given their remarkably conserved phenotypes.
More surprising, however, is the contrasting pattern of morphological evolution in island zosteropids. These birds have explored more novel regions of morphospace, although their phenotypic divergence is again not characterized by an early burst and instead generally not distinguishable from a random-walk model. This pattern may be the result of island clades evolving into a larger morphospace around the generalized bauplan of the 'typical' Zosterops, as opposed to evolving truly novel phenotypic trajectories as observed in some island lineages of birds Lerner et al., 2011;Lovette et al., 2001;Reddy et al., 2012) and lizards (Pinto et al., 2008). We suggest the expanded scope of morphological evolution in insular Zosterops arises from reduced interspecific competition with passerine species in general, as well as with congeners, leading to a relaxation of the evolutionary constraints experienced by continental taxa. Unlike previous studies making broadly similar claims (e.g. , however, we find no evidence for a consistent tendency towards enlarged body sizes predicted by the 'island rule'. Overall, our results suggest that divergent evolutionary pathways on continents and islands are not limited to famed adaptive radiations or 'island rule' effects, but can help to explain more subtle patterns of diversification in the world's 'great speciators'.

ACK N OWLED G EM ENTS
We are grateful to Robert Prŷs-Jones, Mark Adams and Hein Van Grouw for facilitating collection visits to the Natural History Museum at Tring. We also thank Adam Devenish and Catherine Sheard for help with data curation and collection, and Ally Phillimore, Sushma Reddy and two anonymous reviewers for providing helpful comments on an earlier version of this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
Biometric data are available from the DRYAD Digital Repository https://doi:10.5061/dryad.x3ffb g7g3.